ARTICLE

https://doi.org/10.1038/s41467-019-14069-2 OPEN Endogenous FGF21-signaling controls paradoxical resistance of UCP1-deficient mice

Susanne Keipert 1,2,3*, Dominik Lutter 1,2, Bjoern O. Schroeder 4,9, Daniel Brandt1,2, Marcus Ståhlman4, Thomas Schwarzmayr5, Elisabeth Graf5, Helmut Fuchs 6, Martin Hrabe de Angelis 2,6,7, Matthias H. Tschöp 1,2,8, Jan Rozman 2,6,10 & Martin Jastroch1,2,3*

Uncoupling protein 1 (UCP1) executes thermogenesis in brown adipose tissue, which is a

1234567890():,; major focus of human obesity research. Although the UCP1-knockout (UCP1 KO) mouse represents the most frequently applied animal model to judge the anti-obesity effects of UCP1, the assessment is confounded by unknown anti-obesity factors causing paradoxical obesity resistance below thermoneutral temperatures. Here we identify the enigmatic factor as endogenous FGF21, which is primarily mediating obesity resistance. The generation of UCP1/FGF21 double-knockout mice (dKO) fully reverses obesity resistance. Within mild differences in energy metabolism, urine metabolomics uncover increased secretion of acyl- carnitines in UCP1 KOs, suggesting metabolic reprogramming. Strikingly, transcriptomics of metabolically important organs reveal enhanced lipid and oxidative metabolism in specifically white adipose tissue that is fully reversed in dKO mice. Collectively, this study characterizes the effects of endogenous FGF21 that acts as master regulator to protect from diet-induced obesity in the absence of UCP1.

1 Institute for Diabetes and Obesity, Helmholtz Diabetes Center, Helmholtz Zentrum München, German Research Center for Environmental Health (GmbH), 85764 Neuherberg, Germany. 2 German Center for Diabetes Research (DZD), 85764 Neuherberg, Germany. 3 Department of Molecular Biosciences, The Wenner-Gren Institute, Stockholm University, 106 91 Stockholm, Sweden. 4 Wallenberg Laboratory and Sahlgrenska Center for Cardiovascular and Metabolic Research, Institute of Medicine, University of Gothenburg, 413 45 Göteborg, Sweden. 5 Institute of Human Genetics, Helmholtz Zentrum München, German Research Center for Environmental Health (GmbH), 85764 Neuherberg, Germany. 6 German Mouse Clinic, Institute of Experimental Genetics, Helmholtz Zentrum München, German Research Center for Environmental Health (GmbH), 85764 Neuherberg, Germany. 7 Chair of Experimental Genetics, School of Life Science, Weihenstephan, Technische Universität München, 85354 Freising, Germany. 8 Division of Metabolic Diseases, Department of Medicine, Technische Universität, Munich, Germany. 9Present address: Laboratory for Molecular Infection Medicine Sweden (MIMS) and Department of Molecular Biology, Umeå University, 901 87 Umeå, Sweden. 10Present address: Czech Centre for Phenogenomics, Institute of Molecular Genetics of the Czech Academy of Sciences BIOCEV, 252 50 Vestec, Czech Republic. *email: [email protected]; [email protected]

NATURE COMMUNICATIONS | (2020)11:624 | https://doi.org/10.1038/s41467-019-14069-2 | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-14069-2

besity is a major health burden of our society that causes Results Ocomorbidities, such as diabetes, heart disease, and cancer. UCP1 KO mice are protected from DIO at room temperature. Positive energy balance triggers obesity, either by To prevent confounding effects of cold stress in early develop- increased food intake and high-calorie diets, or by decreased ment, wildtype (WT) and UCP1 KO mice were weaned and energy dissipation, or a combination of both1,2. While food intake raised at thermoneutrality and transferred from 30 to 23 °C and preference are mainly regulated via central mechanisms, (standard housing temperature for laboratory mice) at an age of energy dissipation takes place in peripheral organs. Thermo- 10–12 weeks, and normal chow diet was replaced by high fat genic adipose tissue is considered an attractive target for the diet (HFD). In response to the treatment, body weight gain treatment of obesity and diabetes, evidenced by numerous developed significantly lower in UCP1 KO compared to WT recent studies that aim to activate energy dissipation by mice, resulting in significantly lower body weights after twelve increasing heat production3,4. Rodent models established that weeks of HFD feeding (Fig. 1a). Thus, the ‘paradoxical resistance defective brown adipose tissue (BAT) thermogenesis is involved to DIO on HFD fed UCP1 KO mice’21,23 appears to be a robust in the development of obesity, and activation of BAT thermo- and reproducible phenomenon among different independent genesis reduces weight gain5–9. Albeit the presence of BAT in laboratories. adult humans is well-accepted10–13, its contribution to whole body metabolism is a matter of debate14.NeonatalBATmassis Protection from DIO coincides with increased FGF21. To get limited in adults, while a larger proportion has been classified as mechanistic insights on the unexplained phenomenon of obesity beige adipose tissue15,16. The physiological role of beige adipose resistance, we focused on the time-point after three weeks of HFD tissue has not been clarified yet, but could be the target tissue for feeding before changes in body weight occur (Fig. 1b; indicated by therapeutic intervention in humans. In mice, beige fat mass the red arrow). In a second cohort of mice, no differences in body increases when BAT mass is genetically reduced, suggesting weight, body fat or liver triglycerides were detectable between WT compensatory heat production5. The subcutaneous beige fat, and UCP1 KO mice after three weeks of HFD (Fig. 1c, d). Next, however, may not be able to distribute heat as effectively as we focused on FGF21, as FGF21 serum levels in UCP1 KO mice the neonatal BAT, that directly delivers heat to the heart via the are increased at temperatures below thermoneutrality24. While Sulzer’s vein. From the therapeutic perspective, heat generation endogenous FGF21 imposes no effects on metabolic rates and in the subcutaneous part of the body could be advantageous body weight during chow diet feeding24,25, its role for systemic as increased heat dissipation may prevent hyperthermia in metabolism is established under conditions of obesity, where humans. administration of exogenous FGF21 (or FGF21 analogs) leads to Thermogenic adipose-specific mitochondrial uncoupling improved metabolic profiles in mice and humans26,27. After three protein 1 (UCP1) is solely responsible for the canonical weeks of HFD feeding at 23 °C, we detected significantly mechanism of adaptive heat production17 and is expressed in increased serum levels of endogenous FGF21 in UCP1 KO mice both, brown and beige fat18. Activation of UCP1 increases the (Fig. 1e), which coincide with increased FGF21 expression in mitochondrial proton leak, thereby uncoupling mitochondrial BAT and inguinal white adipose tissue (iWAT), but not in the respiration from ATP synthesis19. The acceleration of nutrient liver (Fig. 1f). oxidation increases all exergonic reactions, which results in increased heat output. Notably, obese humans express only minor amounts of UCP1; and thus, UCP1-independent ther- FGF21 entirely controls DIO resistance in UCP1 KO mice. mogenic mechanisms in adipose tissue may bear higher trans- Next, we generated UCP1/FGF21 double knockout (dKO) mice lational potential20. Interestingly, the importance of UCP1- to evaluate the role of FGF21 signaling during HFD-induced body dependent thermogenesis for body weight regulation is mainly weight development in UCP1 KO mice. WT, FGF21 KO, UCP1 supported by a proof-of-principle study of UCP1-knockout KO, and dKO mice were weaned and raised at 30 °C. At the age of (KO) mice at thermoneutrality, which are slightly prone to diet- 10–12 weeks, all mice were transferred to 23 °C and normal chow induced obesity (DIO)9. Paradoxically, UCP1 KO mice are sur- diet was replaced by HFD for either three (Fig. 2a, SAC1) or prisingly resistant to DIO at standard mouse housing tempera- twelve weeks (Fig. 2a, SAC2). Impressively, dKO mice displayed tures of 20–24 °C21–23. Neither the regulation nor the underlying identical body weight development as found for WT and FGF21 molecular mechanisms have been understood yet, labeling the KO mice under HFD conditions, while UCP1 KO mice stayed observation as ‘paradoxical resistance to diet-induced obesity of lean as expected (Fig. 2b). The body weight gain of WT, FGF21 high fat diet fed UCP1 KO mice’23. A comprehensive compen- KO, and dKO mice was caused by the increase in fat mass, but dium on BAT biology stated that there must be secretion of a not fat-free mass (FFM) (Fig. 2c, d), reflected in increased epi- UCP1-independent ‘antiobesity factor’ from BAT, but that ‘such didymal adipose tissue (eWAT) and iWAT mass (Fig. 2e, Sup- an antiobesity factor remains hypothetical’17. Interestingly, the plementary Fig. 1a, b). In contrast, BAT mass of UCP1 KO and underlying UCP1-independent factors must prevent the accu- dKO mice was increased independently of body fat status and mulation of nutrient energy more effectively than UCP1- FGF21, and thus, was solely driven by the lack of UCP1 (Fig. 2e, dependent thermogenesis at room temperature. Supplementary Fig. 1c). Lower liver triglyceride content in UCP1 To identify the controlling factors and molecular networks KO mice was a consequence of the lean phenotype, as no dif- of paradoxical obesity resistance in UCP1-deficient mice, we ferences were visible after three weeks of HFD (Fig. 2f). Glucose apply an array of technologies, ranging from mouse metabolic tolerance was not different between the genotypes (Fig. 2g) but phenotyping to global omics analyses. Strikingly, we discover UCP1 KO mice were more insulin-sensitive, indicated by lower that endogenous fibroblast growth factor 21 (FGF21) is the basal blood glucose, insulin levels, and HOMA-IR (Fig. 2h–j). single key regulator mediating paradoxical resistance to obesity. Systemic lipid metabolism was characterized by lower leptin Consolidating this finding by generating the UCP1-FGF21 levels in UCP1 KO mice after twelve weeks of HFD (Fig. 2k), double knockout (dKO) mouse, transcriptomic profiling of possibly due to lower fat mass (Fig. 2c). While serum free fatty multiple tissues determines FGF21-dependent white adipose acids were not affected (Fig. 2l), increased triglyceride con- tissue remodeling as the site of inefficient energy turnover, with centrations associate to the lack of FGF21 after twelve weeks of networks suggesting enhanced lipid turnover and oxidative HFD (Fig. 2m). Importantly, the paradoxical resistance of UCP1 metabolism. KO mice to DIO appeared only at 23 °C. Maintaining animals

2 NATURE COMMUNICATIONS | (2020)11:624 | https://doi.org/10.1038/s41467-019-14069-2 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-14069-2 ARTICLE

abc 50 WT UCP1 KO 20 WT UCP1 KO 8 6 *** *** 50 WT WT 40 UCP1 KO UCP1 KO 15 40 6 30 4 10 30 4 20 20 Body fat (g) Body weight (g) 2 5 Body weight (g) 10 Body weight gain (g) 2

10 Body weight gain (g) WT UCP1 KO 0 0 0 0 0 3 weeks HFD 3 weeks HFD chow HFD chow HFD 0 2 4 6 8 10 Time (week on HFD)

de fLiver BAT iWAT 3 120 120 WT 1000 WT 3 WT WT WT UCP1 KO ) UCP1 KO UCP1 KO UCP1 KO UCP1 KO –1 100 100 800 ** *** 2 2 80 80 600 60 60

400 Fgf21 1 1 40 40 ***

Liver triglycerides 200 (mg per mg protein) 20 20 Serum FGF21 (pg ml (relative gene expression

0 0 normalized to HPRT and WT) 0 0 0 3 weeks HFD 3 weeks HFD 3 weeks HFD 3 weeks HFD 3 weeks HFD

Fig. 1 UCP1 KO mice display resistance to DIO and increased FGF21 at room temperature. At the age of 10–12 wks, WT and UCP1 KO mice were transferred from 30 °C to room temperature and high fat diet (HFD) or chow diet was offered for 12 wks. a Mean body weight and body weight gain after 12 wks chow or HFD treatment. b Body weight trajectory at 23 °C on HFD. After 3 wks of HFD feeding at 23 °C, c body fat content and body weight gain, d liver triglycerides, e serum FGF21 levels, and f FGF21 gene expression in liver, BAT and inguinal WAT was assessed. a: n = 4/5/6/6 WT chow/UCP1 chow/WT HFD/UCP1 KO HFD; b: n = 6/6WT/ UCP1 KO; c: n = 6/7 – n = 6/6; d: n = 4/5; e, f: n = 6/7). Data are presented as mean ± SEM. Data were analyzed by Student’s t-test **P < 0.01, ***P < 0.001. Source data are provided as a Source Data file. during the twelve weeks of HFD intervention at thermoneutral the caecum samples from the different groups. Analysis at the conditions (30 °C) led to identical body weight and body fat phylum and family level indicated that the microbiota was accumulation in all genotypes (Fig. 2n, o), coinciding with the dominated by Firmicutes (Fig. 3k, Supplementary Fig. 2g). No lack of FGF21 induction in BAT and iWAT (Supplementary differences between the genotypes were detected neither before Fig. 1d, e). Collectively, these data delineate that increased levels the onset of obesity development (Fig. 3i–l) nor after twelve weeks of FGF21 entirely mediate the paradoxical resistance to DIO of of HFD feeding (Supplementary Fig. 2e–h), suggesting that the UCP1 KO mice at room temperature. microbiome has no impact on the lean phenotype of UCP1 KO mice. Furthermore, differences of the bile acid profile were minor Mouse metabolic phenotyping. Next, we performed compre- between genotypes (Fig. 3m–o, Supplementary Fig. 2i, j). Feces hensive mouse metabolic phenotyping to identify the parameters energy content and food assimilation, assessed by bomb calori- of energy metabolism that are responsible for obesity resistance. metry, did not differ significantly between UCP1 KO and dKO Using indirect calorimetry, no genotypic differences in metabolic mice (Fig. 3p, q). rate were found when normalized to fat free mass (Fig. 3a) or when plotted against whole body weight (Fig. 3b). The use of fat free mass eliminates the confounding impact of differences in Urine composition and excreted energy. Metabolic phenotyping inert fat mass (Fig. 2c, Supplementary Fig. 2a, b). Food intake and revealed increased water intake of UCP1 KO mice (Fig. 3d, water intake were normalized to fat free mass (Fig. 3c, d) and Supplementary Fig. 2d), supporting recent reports on induced – given per animal (Supplementary Fig. 2c, d). Food intake was not drinking by FGF21 administration or endogenous FGF2131 33. significantly different between the genotypes. Activity counts Thus, increased endogenous levels of FGF21 impose physiological (Fig. 3e) and body temperature (Fig. 3f) did not differ sig- consequences in UCP1 KO mice. The increased water intake of nificantly between genotypes. In contrast to previous findings23, UCP1 KO mice (Fig. 3d) was corroborated in a second cohort of no shift in respiratory exchange ratio (RER, Fig. 3g) and its mice (Fig. 4a) that also served to collect urine. We discovered that percent relative cumulative frequency analysis (PRCF; Fig. 3h) urine excretion of UCP1 KO mice was increased (Fig. 4a). To were detected. Recent studies associate the protection against investigate whether energy loss may occur via energy-storing obesity development to the cold stress-microbiome axis,28 metabolites in urine as suggested before34, we measured urine affecting bile acid (BA) composition and energy assimilation28–30. energy content using bomb calorimetry showing that excessive The reduction of ambient temperature from 30 to 23 °C repre- energy is not lost via the urine (Fig. 4b). Urine creatinine and sents cold stress for UCP1 KO mice. Thus, we performed a kinetic albumin concentration were not elevated in UCP1 KO mice experiment and analyzed the microbiota and bile acids of mice (Fig. 4c, d), excluding kidney failure as cause for increased water after three weeks (Fig. 3i–o) and twelve weeks of HFD feeding intake, which is in line with previous observations32. While tar- (Supplementary Fig. 2e–j). To investigate how diet and ambient geted quantitative metabolomics reveals no changes in hexose temperature affect the microbiota phylogenetic richness in each concentration (Fig. 4e), an increased secretion of acyl-carnitines caecum sample, we analyzed the alpha -diversity assessed by rar- and amino acids in the urine of UCP1 KO mice was observed, efaction and phylogenetic diversity, followed by principal com- indicative of altered metabolism (Fig. 4f, g; the full list of meta- ponent analysis (PCA) of unweighted UniFrac distances between bolites can be found in Supplementary Table 1).

NATURE COMMUNICATIONS | (2020)11:624 | https://doi.org/10.1038/s41467-019-14069-2 | www.nature.com/naturecommunications 3 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-14069-2

abcd40 15 30 4 Genotypes WT FGF21 KO UCP1 KO UCP1 KO 30 Wildtype FGF21 KO dKO 10 dKO 20 20

30°C 23°C WT 5 10 WT

10 Fat mass (g) Weaned and raised Time on high fat diet Body weight (g) FGF21 KO FGF21 KO UCP1 KO Fat free mass (g) UCP1 KO dKO dKO Change of diet AND 0 0 0 ambient temperature Week 3 Week 12 SAC1 SAC2 0 264 8 10 12 024 68 1012 0 264 8 10 12 Time (week on HFD) Time (week on HFD) Time (week on HFD)

efgc 3 WT 7 WT 400

FGF21 KO 6 FGF21 KO ) ac UCP1 KO UCP1 KO –1 a dKO dKO c 300 5 ac 2 a 4 200 b ac 3 a c b 1 2 100 Tissue weight (g) Liver triglycerides

b (mg per mg protein) b b 1 a a Blood glucose (mg dl WT UCP1 KO FGF21 KO dKO 0 0 0 eWAT iWAT BAT 3 weeks 12 weeks 0 50 100 150 200 Time (min)

hij200 5 25 k30 WT l80 WT UCP1 KO FGF21 KO dKO ) FGF21 KO c ) a ) ac –1 –1

–1 UCP1 KO a 4 20 a a a ) a 150 c dKO 60 –1 20 b 3 ac 15 100 ab ab 40 2 10 b HOMA-IR 10 b b

50 Insulin (ng ml 20 1 5 Serum leptin (ng ml Serum NEFA (mg dl Blood glucose (mg dl 0 0 0 0 0 3 weeks 12 weeks 3 weeks 12 weeks WT dKO WT dKO WT dKO

FGF21UCP1 KO KO FGF21UCP1 KO KO FGF21UCP1 KO KO

m n 50 o 25

) UCP1 KO UCP1 KO 300 WT UCP1 KO WT WT –1 dKO FGF21 KO dKO FGF21 KO dKO FGF21 KO 45 a 20 b a a 200 a b 40 a a 15 a a a a 35 b b b 10

100 30 Body fat (g) b Body weight (g) 25 5

Serum triglycerides (mg dl 0 20 0 3 weeks 12 weeks 23°C 30°C 23°C 30°C

Fig. 2 FGF21 entirely controls the resistance to diet-induced obesity in UCP1 KO mice. a Scheme depicting the study design (SAC: sacrifice). At the age of 10–12 wks, WT, FGF21 KO, UCP1 KO and UCP1/FGF21 dKO mice were transferred from 30 to 23 °C and fed a high fat diet (HFD) for 3 or 12 wks. Trajectories of b body weight, c body fat and d fat free mass. e Weights of epididymal and inguinal WAT; and BAT. f Liver triglycerides after 3 and 12 wks of HFD. g Glucose tolerance test after 8 wks of HFD (4 h food withdrawal); h basal blood glucose, i insulin levels and j HOMA-IR. k Serum leptin, l NEFA and (m) triglyceride (TG) levels after 3 and 12 wks of HFD. n Mean body weight and o body fat of mice fed HFD for 12 weeks at 23 or 30 °C. b–e: n = 9/9/9/9 WT/FGF21 KO/UCP1 KO/dKO; g: n = 8/8/9/9; n: n = 8/7/7/7; I, j: n = 7/7/7/6; f: n = 6/4/4/4 – 9/9/9/9 3 wks WT/FGF21 KO/UCP1 KO/dKO – 12 wks WT/FGF21 KO/UCP1 KO/dKO; k–l: n = 5/4/6/5 – 8/9/9/9; n: n = 9/9/9/9 – 10/9/6/9 23 °C WT/FGF21 KO/UCP1 KO/dKO – 30 °C WT/FGF21 KO/UCP1 KO/dKO; o: n = 9/9/9/8 – 9/9/6/9. Data are presented as mean ± SEM. Data were analyzed by one-way ANOVA followed by Bonferroni test. Different letters indicate significant differences (P < 0.05). Source data are provided as a Source Data file.

Transcriptional profiling reveals FGF21-controlled browning. ablation of FGF21 in the dKO mouse (Fig. 5b), demonstrating To gain mechanistic insights on the molecular basis for changed FGF21-controlled tissue remodeling. Given 912 differentially metabolism and how FGF21 controls obesity resistance, next- expressed (DEGs) in iWAT (Supplementary Fig. 3a), about generation RNA sequencing of all genotypes was performed in 90% of the DEGs were upregulated only in UCP1 KO mice metabolically active tissues (BAT, iWAT, liver, and muscle) after (Fig. 5b). The number of DEGs in liver and muscle were relatively three weeks of HFD (Fig. 5), before the divergence of body minor (Fig. 5c, d, Supplementary Fig. 3a). The Venn analysis weights (Fig. 2b). Hierarchical clustering of significantly regulated of DEGs between UCP1 and dKO mice depicts that FGF21- genes in BAT depicted two distinct gene clusters showing similar mediated gene expression almost exclusively occurs in iWAT expression patterns of WT/ FGF21 KO versus UCP1 KO/ dKO (Fig. 5e). To investigate the involvement of secreted factors other samples (Fig. 5a). Strikingly, massive differential gene regulation than FGF21, we looked at the gene expression of commonly in iWAT of the UCP1 KO mouse was reversed by additional known batokines35,36. Some of these appear to be regulated by the

4 NATURE COMMUNICATIONS | (2020)11:624 | https://doi.org/10.1038/s41467-019-14069-2 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-14069-2 ARTICLE

a bcd 3 0.15 WT 60 WT 0.20 WT FGF21 KO FGF21 KO FGF21 KO UCP1 KO ) UCP1 KO –1 UCP1 KO b dKO dKO 0.15 dKO 2 40 0.10

0.10 a a per g FFM) a

–1 1 20 0.05 Metabolic rate Metabolic 0.05 (kJ d WT UCP1 KO Metabolicrate(kJd FGF21 KO dKO 0 0 0.00 Water intake (g per0.00 g FFM) Food intake (g per g FFM) Day Night 20 25 30 35 40 45 Day Night Day Night Body weight (g)

e f g 0.90 h 100 100 WT 40 WT FGF21 KO FGF21 KO

) UCP1 KO UCP1 KO

) 80 –3 80 dKO 38 2 0.85 dKO /VO 60 2 60 36 0.80 40 40 34 PRCF (%) 0.75 Frequency WT 20 RER (VCO 20 Frequency FGF21 Activity (counts 10 Body temperature (°C) 32 Frequency UCP1 KO Frequency dKO 0 0 0.70 0 Day Night Day Night 0.70 0.75 0.80 0.85 0.90 WT dKO RER (VCO2/VO2) FGF21UCP1 KO KO

i 20 j k 1.0 l 1.0

0.8 0.8 15 0.6 0.6

10 0.4 0.4 Fraction family

Fraction phylum 0.2 0.2 WT

5 PC2 (19.09%) dKO UCP1 KO 0.0 0.0

Phylogenetic diversity WT FGF21 KO UCP1 KO dKO FGF21KO Actinobacteria Proteobacteria 0 Bacteroidetes TM7 Alcaligenaceae Lactobacillaceae PC1 (34.97%) Tenericutes Christensenellaceae RF32 unclassified 0 Firmicutes Clostridiaceae Ruminococcaceae S24-7 WT UCPI KO Clostridiales unclassified Coriobacteriaceae Streptococcaceae 2000 4000 6000 8000 Others 10,000 dKO Erysipelotrichaceae FGF21 KO Lachnospiraceae Sequences per sample m n o 2 WT ** FGF21 KO 10 2.0 ** ** UCP1 KO dKO 1 8 1.5 0 6 1.0 –1 4

Ratio primary/ 0.5 –2 2 Fraction total bile acids [%] Ratio of BA known as secondary bile acids

CA CA FXR agonist/FXR antagonist TCA D LCA HCA 0 0.0 TDCA TLCA aMCAbMCAoMCA CDCAUDCA HDCA TaMCATbMCAToMCA TCDCATUDCA T Conjugated bile acids Unconjugated bile acids W dKO WT dKO P1 KO C UCP1 KO FGF21 KO FGF21U KO p q 5 100

) b ab –1 4 95 ab 3 a 90

2 85

1 80 Feces energy (kJ d Food assimilation (%) 0 0

O WT WT KO dKO 1 dKO

FGF21UCP1 K KO FGF2UCP1 KO

loss of UCP1, but none associates to obesity resistance, as absence of UCP1, is further supported by comparing the tran- their regulation was not differentially affected in dKO mice scriptional profile to a previously published ‘browning sig- (Fig. 5f). Well-established and recently identified ‘browning’ nature’37 (Supplementary Fig. 3b, c). Based on the regulatory markers37,38 were exclusively upregulated in iWAT of UCP1 KO network of the molecular browning signature37, the control over mice (Fig. 5g), and ‘browning’ reprogramming, even in the browning in UCP1 KO mice appears to be predominantly

NATURE COMMUNICATIONS | (2020)11:624 | https://doi.org/10.1038/s41467-019-14069-2 | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-14069-2

Fig. 3 Phenotyping of mouse energy metabolism. After 9 wks of HFD feeding at room temperature, indirect calorimetry was performed with WT, FGF21 KO, UCP1 KO, and UCP1/FGF21 dKO mice. a Energy expenditure per fat free mass (FFM) and b correlation of body weight vs energy expenditure per animal, c day/night food intake per fat free mass, d day/night water intake per fat free mass, e day/night activity, f body temperature and g day/night respiratory exchange ratio (RER) with h percent relative cumulative frequency analysis (PRCF). After 3 wks of HFD feeding the gut microbiota and microbial metabolism were assessed. i Alpha rarefaction plot displaying species richness dependent on sampling depth (max depth = 9410 sequences per sample). j Principal coordinate analysis (PCA, weighted UniFrac) displaying β-diversity of the gut microbial community. The percentage of the variation explained by the plotted principal coordinates is indicated in the axis labels. Each dot represents a caecal community. Relative abundance at k phylum and l family level in caecal community of mice, displaying only relative abundances >1%. m Plasma bile acid (BA) profile and n ratio of primary to secondary BAs, and o ratio of BAs known as FXR agonist/FXR antagonist. p Feces energy content and q percent food assimilation were measured after 6 weeks of HFD feeding. a, b, e, g, h: n = 5/6/6/5 WT/FGF21 KO/UCP1 KO/dKO; c: n = 5/5/6/5, d: n = 4/6/6/5; i–l: n = 5/5/6/4; m, n: n = 6/5/6/7; o: n = 6/5/ 6/5; f, p, q: n = 11/6/12/10. Data (a–g, p, q) were analyzed by one-way ANOVA, followed by Bonferroni test and are presented as mean + SEM. Different letters indicate significant differences (P < 0.05). Data are presented as mean ± SD (i) or mean ± SEM (m–o) with significant differences shown as **P < 0.01. Source data of a–h, m–q, h are provided as a Source Data file.

a b c d 10 WT FGF21 KO 4 15,000 50 b UCP1 KO )

8 ) –1

dKO a –1 40 3 M) µ

–1 6 10,000 a g ml 30 a b ab µ a 2 b

g day 4 ab a 20 a a 5000 2 1 Creatinine ( 10 Albumin ( Urine energy (kJ d 0 0 0 0 Food Water Urine WT dKO WT dKO WT dKO

FGF21UCP1 KO KO FGF21UCP1 KO KO FGF21UCP1 KO KO ef 3000 Amino acids

80 b 100 25 b 20 6 b 2000 c 80 20 b 60 a a ab 15 a 4 ab a 60 a ab 15 ab a ab a a Hexose 40 a 10 1000 40 b 10 Alanine 2 Isoleucine Threonine

20 5 Thryptophan 20 Phenylalanine 5 mol per mmol creatinine) µ ( mol per mmol creatinine) mol per mmol creatinine) mol per mmol creatinine) mol per mmol creatinine) 0 0 0 0 0 mol per mmol creatinine) 0 µ µ µ µ µ ( ( ( ( (

WT dKO WT dKO WT dKO WT dKO WT dKO WT dKO

UCP 1KO UCP1 KO UCP1 KO UCP1 KO UCP1 KO FGF21UCP1 KO KO FGF21 KO FGF21 KO FGF21 KO FGF21 KO FGF21 KO

g Acylcarnitines

C3 C3-DC (C4-OH) C4:1 C5-M-DC C5:1-DC C5-OH (C3-DC-M) 0.15 0.6 0.08 0.8 0.08 0.4 b b b b ab b b ab 0.06 0.6 0.06 0.3 ab 0.10 ab 0.4 ab a a a ab a 0.04 0.4 0.04 a a a 0.2 a a a a 0.05 0.2 a 0.02 0.2 0.02 0.1 mol per mmol creatinine) mol per mmol creatinine) mol per mmol creatinine) mol per mmol creatinine) mol per mmol creatinine) 0.00 0.0 0.00 mol per mmol creatinine) 0.0 0.00 0.0 µ µ µ µ µ µ ( ( ( ( ( (

WT dKO WT dKO WT dKO WT dKO WT dKO WT dKO

FGF21UCP1 KO KO FGF21UCP1 KO KO FGF21UCP1 KO KO FGF21UCP1 KO KO FGF21UCP1 KO KO FGF21UCP1 KO KO

C6 (C4:1-DC) C7-DC C8 C9 C10 C12 0.4 b 0.25 b 0.20 0.3 b 0.15 b 0.10 b b b a 0.20 0.08 0.3 a 0.15 ac c a a a a 0.2 ab 0.10 a ac c 0.15 a a 0.06 0.2 0.10 a a 0.10 0.04 a 0.1 0.05 0.05 0.1 0.05 0.02 mol per mmol creatinine) mol per mmol creatinine) mol per mmol creatinine) mol per mmol creatinine) mol per mmol creatinine) mol per mmol creatinine) 0.0 0.00 0.00 0.0 0.00 0.00 µ µ µ µ µ µ ( ( ( ( ( (

WT dKO WT dKO WT dKO WT dKO WT dKO WT dKO

FGF21UCP1 KO KO FGF21UCP1 KO KO FGF21UCP1 KO KO FGF21UCP1 KO KO FGF21UCP1 KO KO FGF21UCP1 KO KO

Fig. 4 Urine composition and excreted energy. After 6 wks of HFD feeding at room temperature, a food and water intake were determined and urine was collected. b Urine energy content, c urine creatine concentration, d urine albumin concentration, e urine hexoses, f urine amino acids, and g urine acylcarnitines were measured. Data are presented as box-plots, showing the metabolite distribution. The central mark indicates the median with the bottom and top edges of the box indicating the 25th and 75th percentiles, respectively. The whiskers extend to the most extreme data points. a (food and Urine), b, c, f (except Threonine), g: n = 11/6/11/9 WT/FGF21 KO/UCP1 KO/dKO; a (water): n = 10/6/11/9; d: n = 6/6/9/6; e: n = 11/6/10/9; f: n = 11/5/ 10/9; f (Threonine): n = 11/5/10/9. Data were analyzed by one-way ANOVA followed by Bonferroni test with different letters indicating significant differences (P < 0.05). Source data are provided as a Source Data file.

6 NATURE COMMUNICATIONS | (2020)11:624 | https://doi.org/10.1038/s41467-019-14069-2 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-14069-2 ARTICLE

acBAT b iWAT Liverd Muscle

WT dKO WT dKO

FGF21 KOUCP1 KO FGF21 KOUCP1 KO –3 –2 –1 0 1 23

z score

e UCP1 KO vs. dKO iWAT Liver 529 4 Muscle 1 BAT 2 0

1 0 52 12

0 0 2

0 0

1 FGF21KOUCP1 KO WT dKO WT FGF21KOUCP1 dKOKO

f Batokines g Brown fat markers Fgf21 Bmp8b Gdf15 Dio2 Cidea Cox7a1 P2rx5 Aco2 10 15 b 8 15 8 6 5 5 b b b b b b 8 6 b a 4 4 b 6 b 10 a a 4 10 3 6 a a 3 4 a a a 4 a a a 4 a a 2 a 2 a 5 5 2 a 2 a 2 a 1 1 a a 2 a a log expression log expression log expression 0 0 log expression 0 0 0 0 0 0 Alternative thermogenic pathways Epdr1 Adipoq Rbp4 Pm20D1 Ckmt2 Slc6a8 Ryr2 ATP2a2 5 8 6 6 10 5 b 7 a b 10.0 a a 4 4 a 6 8 9.5 6 b b b 3 4 4 b 3 a 9.0 6 a b b 5 4 ab 2 8.5 2 a 4 8.0 2 2 1 4 1 2 2

7.5 log expression log expression log expression 0 log expression 7.0 3 0 0 0 0 0 Glyceroneogenesis and lipid metabolism Vegfa Slit2 Nrg4 Gk Gpd1 Elovl3 Cpt1b Acaca 10 13 10 b 15 15 9 2.0 8 b b 12 8 b 8 a b ab a a 1.5 7 a 10 10 8 a 11 a 6 6 a 1.0 6 a 10 a 4 4 a a a 7 9 a 5 5 0.5 5 a 2 8 2 log expression log expression log expression 6 log expression 0.0 4 0 7 0 0 0

FGF21 KO dKO WT FGF21 KO UCP1 KO dKO WT UCP1 KO

h UCP1 KO vs. dKO

Oxidation-reduction process Tricarboxylic acid cycle 10 Fatty acid metabolic process 8 Metabolic process 6 4 Lipid metabolic process 2 Mito. respiratory chain complex l assembly 0 Aerobic respiration –2 Mitochondrion organization –4 Mito. electron transport, ubiquinol to cyt c –6 ATP biosynthetic process –8 Fatty acid beta-oxidation iWAT –10 Mito. acetyl-CoA biosyth. process from pyruvate –10 –5 0 5100102030 40 log2 FC –log10 p fdr

regulated by PPAR alpha and PGC1 alpha (Supplementary glyceroneogenesis and lipid metabolism were highly induced in Fig. 3d). At least on the transcriptional level, genes of recently UCP1 KO mice (Fig. 5g). suggested UCP1-independent thermogenic pathways, such as The (GO) enrichment analysis of the DEGs in endogenous uncoupler synthesis39, creatine metabolism40 and iWAT uncovers an unbiased, detailed picture of molecular calcium cycling41 were not differentially expressed, while genes of network enrichments in UCP1 KO mice compared to dKO mice,

NATURE COMMUNICATIONS | (2020)11:624 | https://doi.org/10.1038/s41467-019-14069-2 | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-14069-2

Fig. 5 Transcriptional profiling reveals FGF21-controlled browning of inguinal fat. After 3 wks of HFD feeding at room temperature, organs were harvested for transcriptomic analysis. Heat maps illustrate all significantly expressed transcripts (adjusted p-value < 0.01) in a BAT, b iWAT, c liver and d muscle. Colors refer to gene expression z-score (b–d). e Venn diagram of all DEGs between UCP1 KO and dKO mice (adjusted p-value < 0.01). f, g Box plots showing the distribution of log2 gene expression level. The central mark indicates the median, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The whiskers extend to the most extreme data points. f Batokine expression in BAT and in WAT, g marker genes of 'browning' (upper panel), genes associated with UCP1-independent thermogenesis (middle panel) and genes involved in glyceroneogenesis and lipid cycling (lower panel). h GO enrichment analysis of significantly regulated genes (adjusted p-value < 0.01) in iWAT of UCP1 KO vs. dKO. The top 12 regulated pathways are shown. Dot plots display expression of single genes related to the enriched pathways. Dot color refers to log2 FoldChange, while dot size refers to −log10 significance of regulation. Bar plots display significance of pathway enrichment. n = 5/5/5/5WT/FGF21 KO/UCP1 KO/ dKO. Data (f, g) were analyzed by one-way ANOVA followed by Bonferroni test with different letters indicating significant difference (P < 0.05).

showing association of upregulated pathways to mitochondrial (daily food intake) and output (daily energy expenditure, feces oxidation and lipid metabolism (Fig. 5h). The increased energy, and urine energy) (Supplementary Table 2, raw data), we mobilization of lipids is further supported by morphological calculated the net energy balance of each genotype (Supplemen- changes in iWAT, which shows increased multilocularity in tary Table 2, calculated energy balance). As compared to UCP1 UCP1 KO mice (Supplementary Fig. 4a). Notably, the browning KO mice, the other genotypes display positive energy balance of iWAT was specific for UCP1 KO mice, associating browning to based on the combination of all bioenergetic parameters. The the lean phenotype. In contrast to iWAT, BAT transcriptomics surplus of ~1–2 kcal (1 kcal = 4.2 kJ) per day over the 8 weeks revealed enriched pathways very similar in UCP1 KO and dKO period would lead to fat accumulation that minorly differs from mice (Supplementary Fig. 4b), demonstrating that BAT repro- the experimentally assessed fat mass values (Fig. 2c). With the gramming is predominantly driven by the lack of UCP1, with technically challenging measurements of urine energy content, we only minor impact of FGF21. Almost no differences in metabolic assume that almost all nutrient energy is detected, enabling us to pathways between all genotypes were seen in liver and muscle partition 100% nutrient energy into the energy containing/dis- (Supplementary Fig. 4b). Taken together, the global transcrip- sipating modules (Supplementary Table 2). Here, it transpires tomic analyses exclude BAT as contributing organ to the that WT, FGF21 KO, and dKO mice ingest more energy than they development of obesity resistance in UCP1 KO mice and lose, whereas UCP1 KO mice fully dissipate energy intake. highlights iWAT remodeling as driving force. The molecular Importantly, our bioenergetic calculations emphasize that minor classification supported by clustering and enrichment analyses imbalances in daily energy metabolism can accumulate to sub- suggests enhanced metabolic activity of iWAT in UCP1 KO mice, stantial differences in obesity development. This requires con- that has been also annotated as the ‘browning of white fat’. sideration in studies addressing human obesity, which cannot simply be attributed to single changing parameters. In line with Discussion our findings, other studies with comprehensive mouse pheno- The knockout mouse of thermogenic UCP1 provides probably typing supports this notion showing that single parameters such the most extensively studied animal model to investigate anti- as energy intake or expenditure provide only minor differences obesity mechanisms mediated by BAT. However, these mice leading to exessive fat storage42. express the phenomenon of paradoxical resistance to DIO at The UCP1/FGF21 dKO mouse unambiguously pinpoints room temperature, clouding the interpretation on brown fat endogenous FGF21 as primary crucial mediator of obesity resis- significance in obesity on one hand, but also suggesting the tance, and iWAT as the target organ. These observations are existence of alternative pathways of energy loss on the other. coherent with the pharmacology of exogenous FGF21 that does Using UCP1/FGF21 dKO mice, we show that FGF21 mediates the not require UCP1-dependent thermogenesis for beneficial meta- resistance to DIO and the remodeling of iWAT (‘browning’)in bolic effects during obesity43,44. While mild cold is required as the UCP1 KO mice. Therefore, this study solves the molecular trigger of FGF21 release, this hormone fully unfolds its anti- identity of primary endocrine control during UCP1-independent obesity effects under high-fat diet conditions. In previous studies, (previously paradoxical) obesity resistance, opening a window to others showed that BAT becomes a source of FGF21 in the cold45, identify novel approaches for obesity therapy and to distinguish which is further potentiated in UCP1 KO mice24. The require- UCP1-dependent and -independent contributions to obesity ment to maintain body temperature under long-term cold con- resistance. ditions of 5 °C overwrites FGF21 action, seen as FGF21- Given the importance of therapeutically treating the obesity independent browning in UCP1 KO mice25. epidemic, the underlying mechanisms of energy loss in the pre- While previous morphological observations in iWAT of UCP1 sence and absence of UCP1 are indisputably important. While a KO mice suggested browning of iWAT23, this study provides with substantial body of evidence emphasizes the importance of the transcriptomic analysis an important resource for researchers increasing BAT mass and UCP1 expression to counteract investigating UCP1-independent molecular networks of obesity obesity3,4, it should be noted that the UCP1 KO mouse model relevance and the molecular landscape for FGF21-dependent only supports anti-obesogenic effects of BAT in thermoneutral browning. FGF21-dependent intracellular signaling is mediated conditions9. The pioneering metabolic studies on paradoxical by FGFRs in combination with beta klotho46, but other unknown resistance to obesity in UCP1 KO mice were in line with our factors may contribute. While we did not investigate unknown metabolic assessment showing no significant differences in energy factors of intracellular FGF21 signaling, the genetic network metabolism after dietary change21 and after 2 and 17 weeks of underlying browning and metabolic consequences in iWAT were high fat diet feeding23. In our study, however, we completed the addressed in silico to get further insights into potential pathways. bioenergetic assessment by including feces and urine energy The recently published molecular browning signature and its content. Our comprehensive phenotyping enabled us to search regulation, that is based on meta-analysis of more than 100 for differences in energy balance based on the assumption that published data sets37, has been adopted to our data for prediction the murine system obeys the first law of thermodynamics and no of regulatory pathways. This analysis reveals the molecular energy disappears. Given the absolute values of energy intake induction of browning, controlled by PPAR alpha and PGC1

8 NATURE COMMUNICATIONS | (2020)11:624 | https://doi.org/10.1038/s41467-019-14069-2 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-14069-2 ARTICLE alpha, which were more pronouncedly increased than NR4A1, an performed with male WT, FGF21 KO, UCP1 KO and UCP1/FGF12 double KO adrenergically responsive gene (see Supplementary Fig. 3d). (dKO) mice (genetic background C57BL/6J). The breeding strategy to generate the fi Mapping these transcription factors on highly significant DEGs of double KO mice was as follows: The rst breeding step generated heterozygous UCP1/FGF21 KO mice by breeding homozygous UCP1 KO with homozygous iWAT in UCP1 KO mice elucidates routes for the induction of FGF21 KO mice. Afterwards, those mice were mated to generate homozygous mice lipid metabolism genes, such as Cpt1b and Pdk4. This is further for all genotypes, which were again mated and used for the experiment. The animal supported by an unbiased pathway analysis of all transcriptomic experiments complied with all ethical regulations for animal testing and research, data, revealing the FGF21-dependent enrichment of lipid and including animal maintenance and experimental procedures, that the animal welfare authorities of the local animal ethics committee of the state of Bavaria oxidative metabolism only in iWAT, including the typical (Regierung Oberbayern) approved in accordance with European guidelines. browning genes37. These observations are coherent with phar- 47 macological studies administering FGF21 . Thus, enhanced lipid Gene expression analysis. RNA was extracted using Qiazol according to the 47 metabolism may prevent obesity, as suggested for other mouse manufacturer’s instructions (Qiazol Lysis Reagent, Qiagen, Germany). Synthesis of and clinical studies48. cDNA and DNase treatment were performed from 1 μg of total RNA using FGF21-dependent browning in DIO resistant UCP1 KO mice QuantiTect Reverse Transcription Kit (Qiagen, Germany). Quantitative real-time PCR was performed on the ViiA™ 7 Real-Time PCR System (Applied Biosystems, also raises the question on alternative thermogenic mechanisms. USA). The PCR mix contained SybrGreen Master Mix (Applied Biosystems, USA), The simultaneous induction of lipolysis and lipogenesis fuels the cDNA corresponding to 5 ng of RNA, and gene specific primer pairs. Gene idea of lipid futile cycling as an ATP-dependent thermogenic expression was calculated as ddCT using HPRT for normalization. The data are mechanism49–51, which may serve to maintain the lean phenotype shown as values relative to the WT group. The oligonucleotide primer sequences of the UCP1 KO mouse. Other recently suggested mechanisms, are available in Supplementary Table 3. such as creatine and calcium futile cycling40,41, were induced in Indirect calorimetry and metabolic cages. Energy expenditure, food intake, cold-acclimated UCP1 KO mice but the pronounced induction of = activity and respiratory exchange ratio (RER VCO2 produced/VO2 consumed) gene expression was not seen in this study. This is probably not were measured by indirect calorimetry using an open respirometry system with surprising as conditions of mild cold with overexcessive energy simultaneous measurements of activity, food and water intake (TSE PhenoMaster, intake impose less thermoregulatory constraints. TSE Systems, Germany). Measurements were performed in 10 min intervals over a Not obviously affecting energy expenditure, food intake or period of 4 days. The last 2 days were analyzed to assess genotypic differences. Mice were house individually in metabolic cages allowing the monitoring of food and body temperature, the thermogenic potential of FGF21-controlled water uptake, as well as feces and urine collection (Tecniplast, Hohenpeissenberg, browning of iWAT may be weak as compared to BAT, but the Germany). After 2 days of acclimation, 24-h food and water intake were determined metabolic impact appears to be powerful for preventing obesity. by daily weighing of food hoppers and water bottles. Feces and urine were collected Seminal findings by others demonstrated FGF21’s potency to in the same time intervals. Body mass was monitored by daily weighing of mice. 52,53 Urine samples of three 24-h cycles were transferred to Eppendorf tubes, centrifuged directly brown WAT , but whether FGF21 impacts browning at 5000 × g for 10 min. Aliquots for subsequent analysis were stored at −20 °C. also indirectly in the UCP1 KO mouse, remains to be determined. fi Our data reveal no signi cant impact of endogenous FGF21 on Bomb calorimetry. Feces samples (about 1 g each) were dried at 60 °C for two food consumption. In contrast, high doses of exogenous FGF21 days, homogenized in a grinder and condensed with pressure to determine energy reduce food intake of obese UCP1 KO mice43. However, pair- content by bomb calorimetry (IKA C7000, Staufen, Germany). To obtain per- feeding experiments also reveal that this only partially explains cent of food assimilated energy, the energy content of feces was subtracted from 43 energy uptake which was set to 100%. To determine the amount of energy lost via the body weight reduction . In WT mice, increased levels of urine, 1 ml of thawed urine was loaded on cotton coils as combustion aid (Cotton FGF21 are usually associated with increased energy expenditure Cell Rolls 8 mm, Henry Schein Inc., Melville, NY, USA) that had been desiccated to but no differences in food intake, as seen during FGF21 gene constant weight at 60 °C, prior to loading56. The exact volume could be determined 54 55 by using high precision pipettes. The energy content of unloaded cotton coils was therapy , by transgenic overexpression of FGF21 and exo- − 26 16.918 kJ g 1 dry weight. Loaded coils were then carefully handled and desiccated genous FGF21 administration . Recently published effects on again to constant weight at 60 °C in a drying oven. The weight difference between drinking behavior were evident in our metabolic phenotyping, the unloaded dry cotton coil and the loaded dry cotton coil accounted for the dry demonstrating physiological significance of endogenous FGF21. residues contained in the respective volume of loaded urine. The loaded cotton The subsequent higher urine excretion, however, did not reveal coils were then combusted in a bomb calorimeter (IKA C7000, Staufen, Germany). significant impact on energy loss by increased secretion of amino The energy content of the dry urine residue was automatically calculated by the software of the bomb calorimeter, taking into account the total supplemented acid and lipids, but hinted towards altered metabolism in the energy from both the cotton coil and the fuse, which was needed to initiate the UCP1 KO mouse. combustion of the sample. We then normalized urine residue energy content to 1 Beyond the UCP1 KO model, the control of obesity resistance ml of urine. To calculate daily energy loss, normalized urine energy content was by adipose FGF21 may also explain the lack of obesity resistance multiplied by the total amount of excreted urine. in other important BAT-deficient mouse models, such as mice lacking ß-adrenergic receptors (ß-less) or mice with isolated BAT Serum analysis. For the analysis of serum FGF21, Leptin, NEFA, Triglycerides, as fi 6,7 well as Insulin, commercially available assay kits were used according to the de ciency (UCP1-DTA) . In the ß-less mice, the adrenergic manufacturer’s recommendations (Intact FGF-21 ELISA Kit, Eagle Biosciences; stimulation of adipose tissue that stimulates FGF21 release, Mouse Leptin ELISA Kit, Crystal Chem; NEFA-HR2 Wako, LabAssay Triglyceride should be missing7, while in BAT-less mice, there is no FGF21- Wako and Insulin Elisa, Alpco). release from BAT24. Collectively, endogenous FGF21-signaling is fi Liver triglycerides. The triglyceride concentration of liver samples was measured an interesting therapeutic target, as it appears suf cient to prevent − obesity in the absence of UCP1. after extraction with 10 mmol l 1 sodium phosphate buffer (pH 7.4) containing 1 mmol l−1 EDTA and 1% polyoxyethylene (10) tridecylether using a Triglyceride Kit according to the manufacturer’s recommendations (LabAssayTM Triglyceride, Wako). Methods Animals. The experiments were performed in homozygous male WT and UCP1 Glucose tolerance test. After 8 weeks of dietary intervention, an intra-peritoneal KO mice (genetic background C57BL/6J). We reduced confounding developmental (i.p.) glucose tolerance test (GTT) was performed. Glucose (2 mg g−1 of body adaptation to thermal stress by breeding, raising and maintaining all mice at weight) was applied i.p. 4 h after food withdrawal. Glucose levels were measured thermoneutrality before the experimental procedures. Mice were housed in groups before, 15, 30, 60, and 120 min after glucose application. with ad libitum access to food and water, and a 12:12-h dark–light cycle (lights on:7:00 CET). At the age of ~10–12 weeks, mice were randomly assigned to chow (Altromin) or 58% high fat diet (Research diets, D12331) fed groups, and housing Transcriptomic analysis. Total RNA was extracted from inguinal WAT, BAT, temperature was changed from 30 °C to 23 °C+/−1 °C. After 3 weeks and after liver and muscle of WT, FGF21 KO, UCP1 KO and dKO mice (n = 5) using Qiazol 12 weeks of dietary intervention, mice were sacrificed 3–4 h after lights went on, according to the manufacturer’s instructions (Qiazol Lysis Reagent, QIAGEN). The and serum and tissue samples were collected. Identical experiments were quality of the RNA was determined with the Agilent 2100 BioAnalyzer (RNA 6000

NATURE COMMUNICATIONS | (2020)11:624 | https://doi.org/10.1038/s41467-019-14069-2 | www.nature.com/naturecommunications 9 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-14069-2

Nano Kit, Agilent). All samples had an RNA integrity number (RIN) value greater at 20,000 × g, the supernatant was evaporated under a stream of nitrogen, and than 8. For library preparation, 1 µg of total RNA per sample was used. RNA reconstituted in 200 µl methanol:water [1:1]. The samples were injected (5 µl) and bile molecules were poly(A) selected, fragmented, and reverse transcribed with the acids were separated on a C18 column (1.7 µ, 2.1 × 100 mm; Kinetex, Phenomenex, Elute, Prime, Fragment Mix (EPF, Illumina). End repair, A-tailing, adaptor liga- USA), using water with 7.5 mM ammonium acetate and 0.019% formic acid (mobile tion, and library enrichment were performed as described in the TruSeq Stranded phase A) and acetonitrile with 0.1% formic acid (mobile phase B). The chromato- mRNA Sample Preparation Guide (Illumina). RNA libraries were assessed for graphic separation started with 1 min isocratic separation at 20%B. The B-phase was quality and quantity with the Agilent 2100 BioAnalyzer and the Quant-iT Pico- then increased to 35% during 4 min. During the next 10 min the B-phase was Green dsDNA Assay Kit (Life Technologies). Strand-specific RNA libraries were increased to 100%. The B-phase was held at 100% for 3.5 min before returning to sequenced as 100 bp paired-end runs on an Illumina HiSeq4000 platform. The 20%. The total runtime was 20 min. Bile acids were detected using multiple reaction STAR aligner* (v 2.4.2a)57 with modified parameter settings (–twopassMode = monitoring (MRM) in negative mode on a QTRAP 5500 mass spectrometer (Sciex, Basic) was used for split-read alignment against the mouse genome assembly Concord, Canada) and quantified using external standard curves. mm10 (GRCm38) and UCSC known Gene annotation. To quantify the number of reads mapping to annotated genes we used HTseq-count° (v0.6.0). Raw read counts Targeted quantitative urine metabolomics. The urine metabolites were analyzed were normalized and DEG were estimated using R package DESeq2. Genes with a with a targeted quantitative and quality controlled metabolomics approach using maximum expression of 15 counts or below were removed from the data. We the AbsoluteIDQ p180 Kit (Biocrates Life Science AG). The sample preparation further removed genes with a cumulative count sum of 75 or below as well as genes and analysis were performed from Biocrates (Biocrates Life Science AG). This that showed no expression at all in more than 10 samples. Hierarchical clustering validated assay allows the comprehensive identification and the quantification of was performed using ‘Euclidean’ distance measure and nearest distance linkage 186 endogenous metabolites, including 21 amino acids, 19 biogenic amine, 40 ACs, method. Significance of Gene Ontology enrichments were estimated using a 76 phosphatidylcholines (PCs), 14 lysophosphati- dylcholines (lysoPCs), hypergeometrical distribution test. All calculations were done using R version 3.4 15 sphingomyelins, and the sum of hexoses (see complete list of measured amino and Matlab R2018a. acids, biogenic amines and ACs in Supplementary Table 2). Acylcarnitines, (Lyso-) All RNA Seq data have been deposited into the gene expression omnibus (GEO) phosphatidylcholines, sphingomyelins, and hexoses were quantified by FIA- MS/ under the accession number GSE122167. MS using a SCIEX 4000 QTRAP® (SCIEX, Darmstadt, Germany) instrument with electrospray ionization (ESI). Quantification of amino acids and biogenic amines DNA extraction and 16S rRNA gene sequencing. Mouse fecal pellets were col- was based on TEA (triethylamine) derivatization in the presence of internal lected and preserved at −80 °C until processing. DNA extraction and 16S rRNA standards followed by LC-MS/MS using a Waters XEVOTM TQ- Smicro (Waters, sequence analysis, genomic DNA from fecal pellets was extracted by repeated bead- Vienna, Austria). The experimental metabolomics measurement technique is beating using a Fast-Prep System with Lysing Matrix E (MPBio, CA). followed by described in detail by patents EP 1 897 014 B1 and EP 1 875 401 B1 (S.L. Ramsay). ammonium acetate precipitation, and purification by QIAmp DNA mini kit The reported lipid annotation represents a sum signal of all isobaric lipids with (Qiagen, Germany). Bacterial DNA was profiled by sequencing of the V4 region of the same molecular weight (±0.5 Da range) within the same lipid class. Analyzed the 16S rRNA gene on an Illumina MiSeq (llumina RTA v1.17.28; MCS v2.5) using glycerophospholipids are differentiated according to the presence of ester and ether 515F and 806R primers designed for dual indexing58 and the V2 kit (2 × 250 bp bonds in the glycerol moiety. The “aa” indicates that fatty acids are, at the sn-1 and paired-end reads). Samples were amplified in duplicates in reaction volumes of 25 μl sn-2 position, bound to the glycerol backbone via ester bonds, whereas “ae” containing 1x Five Prime Hot Master Mix (Quantabio, MA), 200 nM of each pri- indicates that fatty acids in the sn-1 or at sn-2 position are bound via an ether mer, 0.4 mg ml−1 BSA, 5% DMSO and 20 ng of genomic DNA. PCR was carried out bond. The total number of carbon atoms and double bonds present in the lipid under the following conditions: initial denaturation for 3 min at 94 °C, followed by fatty acid chains are annotated as “C x:y”, where x indicates the number of carbons 25 cycles of denaturation for 45 s at 94 °C, annealing for 60 s at 52 °C and elongation and y the number of double bonds. For sphingomyelins, only fatty acids bound to for 90 s at 72 °C, and a final elongation step for 10 min at 72 °C. Replicates were the glycerol backbone at the sn-2 position are indicated under the assumption that combined, purified with the NucleoSpin Gel and PCR Clean-up kit (Macherey- sphingosine (d18:1) is bound at the sn-2 position. For the FIA-MS/MS analysis, Nagel, Germany) and quantified using the Quant-iT PicoGreen dsDNA kit MRMs are employed using lipid species-specific molecular ion and lipid class (Thermo Fisher Scientific). Equal amounts of purified PCR products were pooled specific fragment ion. Due to the relatively low mass resolution of the triple and the pooled PCR products were purified again using Ampure magnetic pur- quadrupole MS instrument, the detected MRM signal is a sum of several isobaric ification beads (Agencourt, Danvers, MA) to remove short amplification products. lipids within the same class. For example, according to LIPID MAPS data base Illumina paired-end reads were merged using PEAR59, and quality filtered to (http://www.lipidmaps.org/), the signal of PC aa C36:6 can arise from at least 15 remove reads that had at least one base with a q-score lower than 20 and that were different lipid species that have different fatty acid compositions (e.g. PC 16:1/20:5 shorter than 220 nucleotides or longer than 350 nucleotides. Quality filtered reads and PC 18:4/18:2), various positioning of fatty acid’s sn-1 and sn-2 position (e.g. were analyzed with the software package QIIME60 (version 1.9.1). Sequences were PC 18:4/18:2 and PC 18:2/18:4), and different double bond positions in those fatty clustered into operational taxonomic units (OTUs) at a 97% identity threshold using acid chains (e.g. PC(18:4(6Z,9Z,12Z,15Z)/18:2(9Z,12Z)) and PC(18:4 an open-reference OTU picking approach with UCLUST61 against the Greengenes (9E,11E,13E,15E)/18:2(9Z,12Z)). reference database62 (13_8 release). All sequences that failed to cluster when tested against the Greengenes database were used as input for picking OTUs de novo. Statistical analysis. Statistical analyses were performed using Stat Graph Prism 6 Representative sequences for the OTUs were Greengenes reference sequences or (GraphPad Software, San Diego, CA USA). Student’s t-test (unpaired, 2-tailed) cluster seeds, and were taxonomically assigned using the Greengenes taxonomy and or one-way ANOVA and Bonferroni’s multiple comparisons test were used to fi 63 the Ribosomal Database Project Classi er . Representative OTUs were aligned determine differences between the genotypes. Statistical significance was assumed 60 64 using PyNAST and used to build a phylogenetic tree with FastTree , which was at p < 0.05. Statistical significance of WT to other genotypes are denoted by *p < α β used to calculate - and -diversity of samples using Phylogenetic Diversity. Chi- 0.05, **p < 0.01, ***p < 0.001. Statistical differences between the four different fi 65 meric sequences were identi ed with ChimeraSlayer and excluded from all genotypes are indicated by letters. downstream analyses. Similarly, sequences that could not be aligned with PyNAST, singletons, sequences present in the blank extraction control and very low abundant sequences (relative abundance < 0.005%) were also excluded. Reporting summary. Further information on research design is available in To correct for differences in sequencing depth, the same amount of sequences the Nature Research Reporting Summary linked to this article. was randomly sub-sampled for each group of samples (rarefaction; maximum depth depending on sample group). A bootstrap version of Mann-Whitney-U test Data availability was used to compare the genotype-dependent abundance of OTUs at different fi fi fi All RNA sequencing data that support the ndings of this study have been deposited in taxonomical levels; signi cant differences were identi ed after correction for false the National Center for Biotechnology Information Gene Expression Omnibus (GEO) discovery rate. Abundances higher than 1% are displayed on the genus level. and are accessible through the GEO Series accession number GSE122167. Microbiota 16S QIIME was used to compute alpha diversity from rarefied OTU tables and to rDNA gene sequencing results have been deposited in the ENA sequence read archive determine statistical significance at maximum rarefaction level by using a two- with accession number PRJEB28632. The source data underlying Figs. 1a–f, 2b–o, 3a–h, sample t-test and 999 Monte-Carlo permutations. Beta-diversity and weighted 3m–q, h, 4a–g and Supplementary Figs. 1, 2 are provided as a Source Data file. All other unifrac distance matrix were computed with QIIME and statistical significance of relevant data are available from the corresponding author on request. sample groupings was determined by adonis method and 999 permutations. Microbiota 16S rDNA gene sequencing results have been deposited in the ENA sequence read archive with accession number PRJEB28632 (http://www.ebi.ac.uk/ Received: 14 March 2019; Accepted: 10 December 2019; ena/data/view). Published online: 31 January 2020

Serum bile acids analysis. Bile acids were analyzed using ultra-performance liquid chromatography-tandem mass spectrometry (UPLCMS/MS). Briefly, samples (50 µl) were extracted with 10 volumes of methanol, containing deuterated internal References standards (d4-TCA, d4-GCA, d4-GCDCA, d4-GUDCA, d4-GLCA, d4-UDCA, d4- 1. Hill, J. O., Wyatt, H. R. & Peters, J. C. Energy balance and obesity. Circulation – CDCA, d4-LCA; 50 nM of each). After 10 min of vortex and 10 min of centrifugation 126, 126 132 (2012).

10 NATURE COMMUNICATIONS | (2020)11:624 | https://doi.org/10.1038/s41467-019-14069-2 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-14069-2 ARTICLE

2. Romieu, I. et al. Energy balance and obesity: what are the main drivers?. 33. Talukdar, S. et al. FGF21 regulates sweet and alcohol preference. Cell Metab. Cancer Causes Control 28, 247–258 (2017). 23, 344–349 (2016). 3. Trayhurn, P. Origins and early development of the concept that brown 34. Elvert, R. et al. Energy loss via urine and faeces–a combustive analysis in adipose tissue thermogenesis is linked to energy balance and obesity. diabetic rats and the impact of antidiabetic treatment on body weight. Biochimie 134,62–70 (2017). Diabetes Obes. Metab. 15, 324–334 (2013). 4. Scheele, C. & Nielsen, S. Metabolic regulation and the anti-obesity perspectives 35. Deshmukh, A. S. et al. Proteomics-based comparative mapping of the of human brown fat. Redox Biol. 12, 770–775 (2017). secretomes of human brown and white adipocytes reveals EPDR1 as a novel 5. Schulz, T. J. et al. Brown-fat paucity due to impaired BMP signalling induces batokine. Cell Metab. https://doi.org/10.1016/j.cmet.2019.10.001 (2019). compensatory browning of white fat. Nature 495, 379–383 (2013). 36. Villarroya, F., Cereijo, R., Villarroya, J. & Giralt, M. Brown adipose tissue as a 6. Lowell, B. B. et al. Development of obesity in transgenic mice after genetic secretory organ. Nat. Rev. Endocrinol. 13,26–35 (2017). ablation of brown adipose tissue. Nature 366, 740–742 (1993). 37. Cheng, Y. et al. Prediction of adipose browning capacity by systematic 7. Bachman, E. S. et al. betaAR signaling required for diet-induced integration of transcriptional profiles. Cell Rep. 23, 3112–3125 (2018). thermogenesis and obesity resistance. Science 297, 843–845 (2002). 38. Ussar, S. et al. ASC-1, PAT2, and P2RX5 are cell surface markers for white, 8. Ravussin, Y., Xiao, C., Gavrilova, O. & Reitman, M. L. Effect of intermittent beige, and brown adipocytes. Sci. Transl. Med. 6, 247ra103 (2014). cold exposure on brown fat activation, obesity, and energy homeostasis in 39. Long, J. Z. et al. The Secreted Enzyme PM20D1 Regulates Lipidated Amino mice. PloS ONE 9, e85876 (2014). Acid Uncouplers of Mitochondria. Cell 166, 424–435 (2016). 9. Feldmann, H. M., Golozoubova, V., Cannon, B. & Nedergaard, J. UCP1 40. Kazak, L. et al. A creatine-driven substrate cycle enhances energy expenditure ablation induces obesity and abolishes diet-induced thermogenesis in mice and thermogenesis in beige fat. Cell 163, 643–655 (2015). exempt from thermal stress by living at thermoneutrality. Cell Metab. 9, 41. Ikeda, K. et al. UCP1-independent signaling involving SERCA2b-mediated 203–209 (2009). calcium cycling regulates beige fat thermogenesis and systemic glucose 10. Nedergaard, J., Bengtsson, T. & Cannon, B. Unexpected evidence for active homeostasis. Nat. Med. 23, 1454–1465 (2017). brown adipose tissue in adult humans. Am. J. Physiol. Endocrinol. Metab. 293, 42. Olszewski, P. K. et al. Neurobeachin, a regulator of synaptic protein targeting, E444–E452 (2007). is associated with body fat mass and feeding behavior in mice and body-mass 11. Saito, M. et al. High incidence of metabolically active brown adipose tissue in index in humans. PLoS Genet. 8, e1002568 (2012). healthy adult humans: effects of cold exposure and adiposity. Diabetes 58, 43. Samms, R. J. et al. Discrete aspects of FGF21 in vivo pharmacology do not 1526–1531 (2009). require UCP1. Cell Rep. 11, 991–999 (2015). 12. van Marken Lichtenbelt, W. D. et al. Cold-activated brown adipose tissue in 44. Veniant, M. M. et al. Pharmacologic effects of FGF21 are independent of the healthy men. N. Engl. J. Med. 360, 1500–1508 (2009). “browning” of white adipose tissue. Cell Metab. 21, 731–738 (2015). 13. Virtanen, K. A. et al. Functional brown adipose tissue in healthy adults. N. 45. Hondares, E. et al. Thermogenic activation induces FGF21 expression and Engl. J. Med. 360, 1518–1525 (2009). release in brown adipose tissue. J. Biol. Chem. 286, 12983–12990 (2011). 14. Moonen, M. P. B., Nascimento, E. B. M. & van Marken Lichtenbelt, W. D. 46. Ogawa, Y. et al. BetaKlotho is required for metabolic activity of fibroblast Human brown adipose tissue: underestimated target in metabolic disease? growth factor 21. Proc. Natl Acad. Sci. USA 104, 7432–7437 (2007). Biochimica et biophysica acta, https://doi.org/10.1016/j.bbalip.2018.05.012 47. Schlein, C. et al. FGF21 Lowers plasma triglycerides by accelerating (2018). lipoprotein catabolism in white and brown adipose tissues. Cell Metab. 23, 15. Cypess, A. M. et al. Anatomical localization, gene expression profiling and 441–453 (2016). functional characterization of adult human neck brown fat. Nat. Med. 19, 48. Ryden, M., Andersson, D. P., Bernard, S., Spalding, K. & Arner, P. Adipocyte 635–639 (2013). triglyceride turnover and lipolysis in lean and overweight subjects. J. lipid Res. 16. Lidell, M. E. et al. Evidence for two types of brown adipose tissue in humans. 54, 2909–2913 (2013). Nat. Med. 19, 631–634 (2013). 49. Flachs, P. et al. Induction of lipogenesis in white fat during cold exposure in 17. Cannon, B. & Nedergaard, J. Brown adipose tissue: function and physiological mice: link to lean phenotype. Int. J. Obes. 41, 372–380 (2017). significance. Physiological Rev. 84, 277–359 (2004). 50. Guan, H. P. et al. A futile metabolic cycle activated in adipocytes by 18. Keipert, S. & Jastroch, M. Brite/beige fat and UCP1—is it thermogenesis? antidiabetic agents. Nat. Med. 8, 1122–1128 (2002). Biochimica et. Biophysica Acta 1837, 1075–1082 (2014). 51. Rohm, M. et al. An AMP-activated protein kinase-stabilizing peptide 19. Nicholls, D. G. & Locke, R. M. Thermogenic mechanisms in brown fat. ameliorates adipose tissue wasting in cancer cachexia in mice. Nat. Med. 22, Physiological Rev. 64,1–64 (1984). 1120–1130 (2016). 20. Chouchani, E. T., Kazak, L. & Spiegelman, B. M. New advances in adaptive 52. Fisher, F. M. et al. FGF21 regulates PGC-1alpha and browning of white thermogenesis: UCP1 and beyond. Cell Metab. https://doi.org/10.1016/j. adipose tissues in adaptive thermogenesis. Genes Dev. 26, 271–281 (2012). cmet.2018.11.002 (2018). 53. Ost, M. et al. Muscle mitochondrial stress adaptation operates independently 21. Anunciado-Koza, R., Ukropec, J., Koza, R. A. & Kozak, L. P. Inactivation of of endogenous FGF21 action. Mol. Metab. 5,79–90 (2016). UCP1 and the glycerol phosphate cycle synergistically increases energy 54. Jimenez, V. et al. FGF21 gene therapy as treatment for obesity and insulin expenditure to resist diet-induced obesity. J. Biol. Chem. 283, 27688–27697 resistance. EMBO Mol. Med. https://doi.org/10.15252/emmm.201708791 (2008). (2018). 22. Kozak, L. P. & Anunciado-Koza, R. UCP1: its involvement and utility in 55. Zhang, Y. et al. The starvation hormone, fibroblast growth factor-21, extends obesity. Int. J. Obes. 32, S32–S38 (2008). lifespan in mice. eLife 1, e00065 (2012). 23. Liu, X. et al. Paradoxical resistance to diet-induced obesity in UCP1-deficient 56. Elvert, R. et al. Energy loss via urine and faeces—a combustive analysis mice. J. Clin. Investig. 111, 399–407 (2003). indiabetic rats and the impact of antidiabetic treatment on body weight. 24. Keipert, S. et al. Genetic disruption of uncoupling protein 1 in mice renders Diabetes Obes. Metab. 15, 324–334 (2013). brown adipose tissue a significant source of FGF21 secretion. Mol. Metab. 4, 57. Anders, S., Pyl, P. T. & Huber, W. HTSeq–a Python framework to work with 537–542 (2015). high-throughput sequencing data. Bioinformatics 31, 166–169 (2015). 25. Keipert, S. et al. Long-term cold adaptation does not require FGF21 or UCP1. 58. Kozich, J. J., Westcott, S. L., Baxter, N. T., Highlander, S. K. & Schloss, P. D. Cell Metab. 26, 437–446 e435 (2017). Development of a dual-index sequencing strategy and curation pipeline for 26. Coskun, T. et al. Fibroblast growth factor 21 corrects obesity in mice. analyzing amplicon sequence data on the MiSeq Illumina sequencing Endocrinology 149, 6018–6027 (2008). platform. Appl. Environ. Microbiol. 79, 5112–5120 (2013). 27. Gaich, G. et al. The effects of LY2405319, an FGF21 analog, in obese human 59. Zhang, J., Kobert, K., Flouri, T. & Stamatakis, A. PEAR: a fast and accurate subjects with . Cell Metab. 18, 333–340 (2013). Illumina Paired-End reAd mergeR. Bioinformatics 30, 614–620 (2014). 28. Zietak, M. et al. Altered microbiota contributes to reduced diet-induced 60. Caporaso, J. G. et al. PyNAST: a flexible tool for aligning sequences to a obesity upon cold exposure. Cell Metab. 23, 1216–1223 (2016). template alignment. Bioinformatics 26, 266–267 (2010). 29. Worthmann, A. et al. Cold-induced conversion of cholesterol to bile acids in 61. Edgar, R. C. Search and clustering orders of magnitude faster than BLAST. mice shapes the gut microbiome and promotes adaptive thermogenesis. Nat. Bioinformatics 26, 2460–2461 (2010). Med. 23, 839–849 (2017). 62. DeSantis, T. Z. et al. Greengenes, a chimera-checked 16S rRNA gene database 30. Zietak, M. & Kozak, L. P. Bile acids induce uncoupling protein 1-dependent and workbench compatible with ARB. Appl. Environ. Microbiol. 72, thermogenesis and stimulate energy expenditure at thermoneutrality in mice. 5069–5072 (2006). Am. J. Physiol. Endocrinol. Metab. 310, E346–E354 (2016). 63. Wang, Q., Garrity, G. M., Tiedje, J. M. & Cole, J. R. Naive Bayesian classifier 31. Camporez, J. P. et al. Cellular mechanisms by which FGF21 improves insulin for rapid assignment of rRNA sequences into the new bacterial taxonomy. sensitivity in male mice. Endocrinology 154, 3099–3109 (2013). Appl. Environ. Microbiol. 73, 5261–5267 (2007). 32. Song, P. et al. The hormone FGF21 stimulates water drinking in response to 64. Price, M. N., Dehal, P. S. & Arkin, A. P. FastTree 2–approximately maximum- ketogenic diet and alcohol. Cell Metab. 27, 1338–1347 e1334 (2018). likelihood trees for large alignments. PloS ONE 5, e9490 (2010).

NATURE COMMUNICATIONS | (2020)11:624 | https://doi.org/10.1038/s41467-019-14069-2 | www.nature.com/naturecommunications 11 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-14069-2

65. Haas, B. J. et al. Chimeric 16S rRNA sequence formation and detection in Competing interests Sanger and 454-pyrosequenced PCR amplicons. Genome Res. 21, 494–504 M.H.T. serves as SAB member for Novo Nordisk and ERX Pharmaceuticals Inc. All other (2011). authors declare no competing interests.

Additional information Acknowledgements Supplementary information is available for this paper at https://doi.org/10.1038/s41467- We would like to thank Ann-Elisabeth Schwarz and Sandy Lösecke (Helmholtz Center) 019-14069-2. for excellent technical assistance, and Claudia Stöger for the project management of the GMC analysis. Furthermore, we would like to thank Valentina Tremaroli and Rozita Correspondence and requests for materials should be addressed to S.K. or M.J. Akrami (University of Gothenburg) for bioinformatics assistance processing the micro- biota sequencing data. This work was supported partly by funding from the German Peer review information Nature Communications thanks the anonymous reviewers for Diabetes Center (DZD) to S.K., and the Swedish Research Council (2018-02150 to S.K. their contribution to the peer review of this work. Peer reviewer reports are available. and 2018-03472 to M.J.). M.H.T. is supported by the Helmholtz Portofolio Program “Metabolic Dysfunction”, the Alexander von Humboldt Foundation and the Helmholtz Reprints and permission information is available at http://www.nature.com/reprints Alliance ICEMED-Imaging and Curing Environmental Metabolic Diseases. M.H.d.A. is supported by the German Federal Ministry of Education and Research (Infrafrontier grant Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in 01KX1012), and the German Diabetes Center (DZD). The computations for microbiota published maps and institutional affiliations. composition analyses were performed on resources provided by the Swedish National Infrastructure for Computing (SNIC) through Uppsala Multidisciplinary Center for Advanced Computational Science (UPPMAX). B.S. is supported by a Marie Curie Intra European Fellowship and a Human Frontier Science Program Long-Term Fellowship. Open Access This article is licensed under a Creative Commons Open access funding provided by Stockholm University. Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Author contributions Commons license, and indicate if changes were made. The images or other third party S.K. and M.J. conceptualized the project, designed and performed experiments, inter- material in this article are included in the article’s Creative Commons license, unless preted data, and wrote the paper. D.Lu. performed bioinformatical analyses of RNASeq indicated otherwise in a credit line to the material. If material is not included in the data, and interpreted the data with S.K. B.S. designed, performed and interpreted the article’s Creative Commons license and your intended use is not permitted by statutory microbiome analysis and M.S. performed bile acid serum profile. D.La. performed animal regulation or exceeds the permitted use, you will need to obtain permission directly from experiments and molecular analysis. T.S. and E.G. performed RNA sequencing. J.R. the copyright holder. To view a copy of this license, visit http://creativecommons.org/ performed measurements with metabolic cages for urine/feces analyses, and interpreted licenses/by/4.0/. the feces energy content. J.R. and H.F. designed the GMC study. M.H.T. and M.H.d.A. contributed reagents/materials/analytic tools. All authors critically reviewed and edited the paper and approved the final version of the paper. © The Author(s) 2020, corrected publication 2021

12 NATURE COMMUNICATIONS | (2020)11:624 | https://doi.org/10.1038/s41467-019-14069-2 | www.nature.com/naturecommunications